Boston University
Fusion model using unsupervised and self-supervised learning for lung ultrasound video classification of consolidations
Abstract
dc:description.abstractLung ultrasound (LUS) has emerged as a promising imaging modality for pediatric pneumonia diagnosis due to its portability and safety. Automating LUS interpretation is challenging due to the variability of the data and limited availability of expert annotations. This thesis explores approaches to automate LUS interpretation with minimal reliance on labeled datasets. First, unsupervised methods using hand-crafted and deep features were explored for clustering frame-level patterns. Building on the insights of this study, the thesis then contributes a fusion-based framework that combines self-supervised contrastive learning with unsupervised anomaly detection. The proposed model learns discriminative representations from unlabeled video frames and integrates them with latent-space deviation scores obtained via dimensionality reduction. This hybrid architecture enables patient-independent, explainable classification of LUS videos. Systematic validation experiments demonstrate that the proposed approach has the potential to support real-world clinical decision-making, especially in resource-limited settings where expert interpretation is scarce.
Degree
thesis:*- Name thesis:degree_name
- Master of Science
- Level thesis:degree_level
- masters
- Discipline thesis:degree_discipline
- Computer Science
- Grantor
- Boston University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bhattacharjee, Saunak
- Advisor dc:contributor.advisor
-
- Betke, Margrit
- ORCID dc:identifier.orcid
Subjects
dc:subject × 6Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/2144/53409